Application Research On Matting Algorithm Based On Kinect

With the development of multimedia technology and the technicalization of digital media industry,digital matting and composition,which have a wide range of applications in movies,games and advertising production,as well as medical health care and space detection,become more and more important and are hot topics in research of digital image processing field.Digital image matting technology is used to extract the interesting foreground with transparency.Most traditional natural image matting algorithm requires the user to provide a trimap or other auxiliary information as a reminder,which costs a lot of time and manpower,making it hard to be applied to the video matting.With the development of the acquisition technology of depth information,a series of RGBD matting algorithms combine the depth information are proposed.In this thesis,we mainly studied the natural image matting algorithm based on depth information captured by Kinect,and realized the automatic Kinect matting system without any manual auxiliary information.The main achievements of this thesis are listed as follows:(1)This thesis studied the origin,development and application of Kinect sensor,compared and summarized the difference of hardware configuration and depth image acquisition technology between Kinect v1 and Kinect v2 sensor.To deal with the error in the depth image captured by Kinect v2,we combined the depth information with the guided filter as a RGB-D filter.What’s more,we obtained the depth information iteratively to improve its precision.(2)We made the analysis and comparison of several traditional natural image matting technology,especially the matting algorithms based on the depth information captured by TOF camera.This thesis combined the traditional RGB image matting algorithm with the depth information captured by Kinect sensor,and chose Shared matting as the main algorithm to ensure the real-time and accuracy of this image matting algorithm.(3)We realized the automatically generation method of trimap,and combined the depth information with Shared matting,proposed a regional expansion method based on depth information.This thesis achieved the automatic image matting algorithm based on depth information captured by Kinect sensor,which can perform in real-time and robustly.(4)This thesis designed and achieved a visual real-time matting and composition system based on Kinect.To deal with the difference of tonal and light conditions between foreground person and new virtual background,we achieved the color transfer algorithm of Erik Reinhard in the system.This system can accurately extract the person from the real background and compose it with the new virtual background in real time,which has the same tonal with the new background.And we can save the new composed image finally.